1 s2.0 S1568494619306738 Main

Applied Soft Computing Journal 86 (2020) 105892
Contents lists available at ScienceDirect
Applied Soft Computing Journal

journal homepage: www.elsevier.com/locate/asoc
A multi-objective robust possibilistic model for technology portfolio

optimization considering social impact and different types of financing
∗
Marzieh Shaverdi a , Saeed Yaghoubi b , , Hamidreza Ensafian b
a
School of Progress Engineering, Iran University of Science and Technology, Tehran, Iran
b
School of Industrial Engineering, Iran University of Science and Technology, Tehran, Iran
article info a b s t r a c t
Article history: With respect to limited financial resources, prioritization of technology fields in order to be supported
Received 2 January 2019 financially is a matter of paramount significance that governmental organizations, such as ‘‘Technology
Received in revised form 28 June 2019 Development Funds (TDFs)’’, face with. Innovation and technology development, as the cornerstone of
Accepted 21 October 2019
the economic development of countries, requires making decisions in terms of assigning the best-suited
Available online 25 October 2019
form of financial resources mainly by governments. Accordingly, this study addresses a multi-objective
Keywords: portfolio optimization problem in a multi-period setting with the aim of maximizing the created
Technology portfolio selection jobs – as a key factor in social welfare – as well as intended profit while minimizing the risk of
Financial resource allocation inappropriate portfolio selection. To formulate the proposed mathematical model, different financing
Multi-objective robust possibilistic methods, technology readiness levels (TRL), and return on investment (ROI) associated with each
programming
technological project are taken into account. Afterward, to deal with the uncertainty arisen from
Fuzzy programming
fuzzy parameters, the Multi-Objective Robust Possibilistic Programming approach (MORPP) is applied,
Augmented epsilon-constraint
the performance of which is examined under several computational tests. Finally, to illustrate the
performance of the proposed model and its applicability in practice, the computational results are
shown through a real case study in Iran Innovation & Prosperity Fund (IIPF). The results show that
selecting small and medium-sized enterprises (SMEs) for being financed, is the best option when
increasing job creation is considered in portfolio optimization. Furthermore, the comparison of the
MORPP model results with the deterministic model shows that the solutions obtained from the robust
possibilistic approach outweighed the deterministic model.
© 2019 Elsevier B.V. All rights reserved.
1. Introduction forms of financial support is a challenging issue which govern-

ments usually face with, different methods of which may be
The ever-changing rate at which the technologies are devel- applied during different stages of firms’ life cycle. In fact, tech-
oping and also the increasing complexity of new technologies nology financing has a crucial role in supporting high-tech firms
in the dynamic business environment have made the optimum at different stages of the innovation process. At the first stage of
technology selection much more complicated. Managing the in- firms’ life cycle, the investment is usually accompanied by a high
novation and technology development is mostly accomplished in risk of failure, and private investors are unwilling to finance firms’
a coherent and integrated system, called ‘‘National Innovation activities. Therefore, start-up firms often face with difficulties
System’’ (NIS). NIS consists of institutions, private and public in access to financial resources from banks or other financial
firms, universities, and government agencies, whose goal is to institutions [3]. At the next stages, with the financial growth of
produce, promote, and exploit the knowledge within the national a company, other investors, including governmental ‘‘Technology
borders of countries [1]. ‘‘Innovation Financing System’’ (IFS) is Development Funds’’ (TDFs) and private investors, may provide
one of the subsystems of NIS whose primary responsibility is financial resources for such firms. Finally, when firms become
financing the countries’ technological development. Technology mature, they may have the opportunity of enjoying Initial Pub-
financing could be defined as the process of funding the in- lic Offerings (IPO) and stock markets to finance their activities
novative businesses to turn their technological inventions into [2]. In this study, the technological projects can enjoy different
commercial innovative products or services [2]. Financing the ways of financing: two types of loan – long-term and short-term
innovative firms’ technological activities with the appropriate loans – joint venture, and direct investment. The decisions as to
determining how the technological projects could be financed
∗ Corresponding author. and how many financial resources should be allocated to each
E-mail address: [email protected] (S. Yaghoubi). project depend on manifold factors; among which the return on
https://doi.org/10.1016/j.asoc.2019.105892
1568-4946/© 2019 Elsevier B.V. All rights reserved.
2 M. Shaverdi, S. Yaghoubi and H. Ensafian / Applied Soft Computing Journal 86 (2020) 105892
investment, the risk, and TRL associated with each project are 2. Literature review
regarded in the proposed mathematical modeling.
TDFs are governmental financial institutions which have the In this section, the most related literature in terms of ‘‘project
responsibility of supporting technology development by attract- portfolio selection’’, ‘‘technology portfolio management’’,
ing the contribution of the private sector. Since the financial ‘‘resource allocation’’, and ‘‘risk management’’ is reviewed. Then
resources are limited, the TDFs should have a prioritization mech- the related literature about robust programming is studied.
anism to optimize the allocation of resources to different technol-
ogy fields and various firms executing the technological projects. 2.1. Technology portfolio
Regarding the importance of technological innovation and
‘‘knowledge-based economy’’, some studies have explored the Technology portfolio management includes a wide range of
concept of the technology financing and commercialization sys- tools and selection methods ranging from quantitative to qualita-
tems in different countries [2–4], which have depicted the big tive measures. Noteworthy, this section mainly focuses on studies
picture of innovation financing policies, IFS of countries, their with the mathematical framework and quantitative models. The
institutions, and the role of them in each phase of technology most related studies are briefly shown in Table 1.
development. However, there are limited studies investigating Technology projects, as the key elements to keep firms com-
technology portfolio optimization and optimal resource allocation petitive in the business world, need to be continually optimized
by decision makers in order to allocate limited resources to a
in TDFs, motivating this study to focus on the role of TDFs in
subset of possible projects which bring about the most profit and
IFS, as organizations that finance the countries’ innovation and
lowest risk. To this end, a variety of tools and methods can be
technology development.
applied to select the optimal set of technology projects. In order
Considering the aforementioned concerns, and also by bearing
to optimize a portfolio of product development improvement
in mind that different technologies have different effects on the
projects, Dickinson et al. [5] using a dependency matrix, pre-
economy, job creation, and country’s overall position at innova-
sented a nonlinear integer model developed for the Boeing Com-
tion and competitiveness rankings, optimal resource allocation
pany, which quantifies the interdependencies between projects.
can enhance the productivity and efficiency of the TDFs’ finan-
In addition, their model seeks to balance the risk, cost, and benefit
cial resources significantly. With this in mind, in this paper, a
of the entire portfolio.
mathematical framework for TDFs’ technology portfolio selection In his book entitled ‘‘Technology Portfolio Planning and Man-
is developed which makes it possible to determine the optimal agement’’, Yu [6] developed a linear single-period model in which
technology portfolio composition and the appropriate type of the limitations of the budget, employee’s time, and managerial
financing. In doing so, the decisions are made based on the time are incorporated into modeling while considering uncertain
corresponding technology fields’ risk, return on investment (ROI), conditions. Considering enterprise benefits – including economic
job creation, and technology readiness level (TRL). and environmental benefits – and synergies between different
The main contributions of this study that differentiates it from technologies, Congbo et al. [7] proposed a methodology to op-
relevant studies are as follows: timize a green technology portfolio. They used the AHP method
to calculate the weights of the enterprise benefits and the ANP
• Taking into consideration the social and economic impact method to assess the synergies among the different types of the
of each technology field to select the optimal technology technologies.
portfolio Since traditional technology assessment tools are often inca-
• Considering the risk, technology readiness level (TRL), tech- pable to consider vague data gathered from a real-world setting,
nology complexity and technology leverage factor (TLF) for Sattari Ardabili [8] developed a multi-objective, single-period
technology portfolio optimization model with random fuzzy technologies’ return in order to op-
• Considering different types of financing methods (i.e. differ- timize a technology portfolio selection problem. To successfully
ent loan types, joint venture, and direct investment) to fi- tackle investment projects in renewable energies, Davoudpour
nance the technological projects according to their financial et al. [9] presented a mathematical model for renewable tech-
requirements, TRL, and ROI nology portfolio selection applied in the oil industry R&D center
• Considering two levels for technology portfolio planning, with the aim of maximizing support of the organization’s strat-
that is ‘‘technology fields’’ and ‘‘technological projects’’ cat- egy and values. Besides, their model tries to set an appropriate
egorized in each technology field balance between cost and benefit of the entire portfolio. Using
• Applying a multi-objective robust possibilistic model to cope AHP method, they first determined the importance of assessment
with uncertain parameters and investigating the behavior of criteria of renewable technology portfolio; and then, proposed
the robust model under several computational tests a mathematical model with deterministic variables for portfolio
• Demonstrating the applicability of the proposed model optimization problem.
through a real case study developed in Iran’s Technology The difficulty of selecting an appropriate set of portfolios is
Development Fund (TDF) mainly resulted from the multiple and often conflicting objec-
tives, inherent technical complexities, and valuation uncertainties
The remainder of this paper is organized as follows: the related associated with the assessment process. Tavana et al. [10] pro-
literature is reviewed in Section 2. In Section 3, the proposed posed a data envelopment analysis (DEA) model addressing the
mathematical model for technology portfolio selection is formu- ambiguity and vagueness of parameters. Their multi-objective
lated. In Section 4, the augmented epsilon constraint approach fuzzy linear model was developed in order for assessing high-
and the multi-objective robust possibilistic approach are incor- tech projects in National Aeronautics and Space Administration
porated into modeling. In Section 5, the real case under investi- (NASA). The proposed model combines different decision-makers
gation is introduced and managerial insights obtained from the opinions about fuzzy inputs and outputs with involving the α -
computational results are presented. Section 5.4 investigates the level-based approach for their fuzzy DEA model. Hassanzadeh
optimality of the applied robust possibilistic approach through et al. [11] suggested a multi-objective model for R&D projects
numerical testing. Section 6 presents the final remarks and future portfolio selection in an uncertain condition. They have used the
research opportunities. robust optimization approach to deal with uncertainty while an
Table 1
Selected resources related to project portfolio and technology portfolio optimization.
Authors
Row
Variables Time Problem type Modeling methods Factors considered in model Application context
M. Shaverdi, S. Yaghoubi and H. Ensafian / Applied Soft Computing Journal 86 (2020) 105892
type period
Uncertain variables
TP/PP at national organization

Different types of financing
MILP/ linear programming
TP/PPa at Company level

Nonlinear programming
Technology complexity
Deterministic variables
Stock market portfolio

Robust programming
Scenario planning
Single-objective
Fuzzy variables
Multi-objective
Single-period
Interval data
Real options
Multi-period
Job creation
TP at TDFb
Stochastic
MCDM
Return
Cost
DEA
Risk
TRL
1 Dickinson et al. [5] ✔ ✔ ✔ ✔ ✔ ✔ ✔
2 Yu [6] ✔ ✔ ✔ ✔ ✔ ✔ ✔ ✔ ✔ ✔
3 Sattari Ardabili [8] ✔ ✔ ✔ ✔ ✔ ✔ ✔ ✔
4 Davoudpour et al. ✔ ✔ ✔ ✔ ✔ ✔ ✔
[9]
5 Tavana et al. [10] ✔ ✔ ✔ ✔ ✔ ✔
6 Hassanzadeh et al. ✔ ✔ ✔ ✔ ✔ ✔ ✔ ✔
[11]
7 Jafarzadeh et al. ✔ ✔ ✔ ✔ ✔ ✔ ✔
[12]
8 Tavana et al. [13] ✔ ✔ ✔ ✔ ✔ ✔ ✔ ✔ ✔ ✔
9 Kocadagli and ✔ ✔ ✔ ✔ ✔ ✔ ✔
Keskin [14]
10 Mashayekhi and ✔ ✔ ✔ ✔ ✔ ✔
Omrani [15]
11 Alvarado et al. [16] ✔ ✔ ✔ ✔ ✔ ✔
12 Mokhtarzadeh ✔ ✔ ✔ ✔ ✔ ✔
et al. [17]
13 Schaeffer and ✔ ✔ ✔ ✔ ✔ ✔
Cruz-Reyes [18]
14 Karasakal and Aker ✔ ✔ ✔ ✔ ✔ ✔ ✔ ✔
[19]
15 Kucukbay and Araz ✔ ✔ ✔ ✔ ✔ ✔ ✔
[20]
16 This paper ✔ ✔ ✔ ✔ ✔ ✔ ✔ ✔ ✔ ✔ ✔ ✔ ✔
a TP: Technology Portfolio, PP: Project Portfolio.

b TDF: Technology Development Fund.
3
interactive procedure has been used in setting tradeoffs among estimation models and five assignment frameworks based on
the multiple objectives. Metrics such as Technology Readiness the DEA approach to sort and rank the R&D projects. They used
Level (TRL) are handled to specify the level of maturity for a given five-level hierarchical structure to define R&D project selection
technology, thereby providing an evaluation of the needed steps criteria. Then, this structure was used to determine the impor-
to motivate the technology implementation. tance of the criteria. Since the evaluation criteria include the
Terrile et al. [21] considered TRL as a proxy for the sake of inputs and outputs of the projects, they developed a DEA-based
reducing the risk of portfolio selection. Likewise, they proposed model to sort the projects.
other measures such as ‘‘Technology Leverage Factor’’ (TLF) and
market size as a proxy for reward in technology portfolio selec- 2.2. Robust optimization
tion. Their model is a qualitative framework developed based on
the risk and the reward matrix. Regarding organization’s goal and There have been developed different methods in order to deal
mission, Tavana et al. [13] considered projects portfolio matching with uncertainty in data. In the cases that the distributional infor-
and proposed a three stages model for projects portfolio selection. mation about random data is available, stochastic programming
They used the DEA method for primary projects screening, the methods are used. In the absence of information about the distri-
TOPSIS for projects ranking, and the linear-integer programming bution of random data, robust programming methods are applied.
for selecting the most appropriate portfolio in a fuzzy setting. Pishvaee et al. [24] presented a robust optimization approach in
Considering the fact that projects have different durations and
order to deal with uncertainty in closed-loop supply chain net-
potential rates of return, and also by taking into account that
work. Their model has a single objective of minimizing total cost
the profit yielded by the implemented projects can be reinvested
of supply chain. Mirzapour Al-e-hashem et al. [25] used robust
in the accomplishment of other projects, Jafarzadeh et al. [12]
optimization approach for a multi-objective aggregate production
used reinvestment strategy within a flexible time horizon and
planning problem.
proposed a model for optimal selection of project portfolios.
The other type of uncertainty is epistemic that is about im-
They applied a three-objective integer programming to determine
precise parameters and arise from the lack of knowledge about
project composition, schedule of the selected projects, and the
the exact values of parameters. In such cases, possibilistic pro-
time horizon of projects’ execution.
gramming approaches are applied. Pishvaee et al. [26] proposed
The multi-objective fuzzy model for portfolio selection pro-
a credibility-based fuzzy programming model for a two-objective
posed by Kocadagli and Keskin [14] took into consideration the
green logistics network. Pishvaee et al. [27] proposed a robust
risk priorities according to market-moving trends as well as the
possibilistic programming for socially responsible supply chain
risk-return tradeoff, which enables decision maker(s) to deter-
network design. They used ‘‘realization model’’ in order to com-
mine the importance and the priority of the designated goals.
pare their model results with the deterministic counterpart and
Kucukbay and Araz [20] presented a linear physical programming
show their model effectiveness. Zahiri et al. [28] presented a
(LPP) model in order to determine the optimal portfolio selec-
multi-period robust possibilistic programming approach for
tion, considering two criteria of expected return and risk. They
location–allocation of organ transplant centers. Zahiri et al. [29]
compared the portfolio selected by applying LPP with the com-
bination of portfolio gained from fuzzy goal programming (FGP) proposed a multi-objective network design for the organ trans-
and found that LPP could be considered a good alternative for plant chain under uncertainty and handled epistemic uncertainty
FGP method. Canbaz and Marle [22] proposed a portfolio selection of data by an interactive fuzzy approach. Salehi Sadghiani et al.
model based on constraint satisfaction problem techniques. Their [30] developed a model in order to design retail supply chain
model considers project interdependencies and balance. network, in which random and possibilistic data was handled by
Schaeffer and Cruz-Reyes [18] proposed a multi-objective math- a robust-possibilistic programming.
ematical model to select R&D projects at a public organization. To the best of our knowledge, none of the studies published
In their framework, each project proposal includes tasks with in the field of technology portfolio optimization integrate the
different types of the costs and financial resources. This frame- technology readiness level (TRL), complexity, technology leverage
work considers interdependencies and synergies among projects factor (TLF) and robust programming in their models. In addi-
and activities. Caglar and Gurel [23] proposed a two-stage model tion, considering different types of financing methods – including
considering the impact of R&D resource allocation to different different loan types, joint venture, and direct investment – with
sectors and the relative balance among them. In the first stage the aim of financing the technological projects is another re-
of the proposed model, sectoral budget allocation decisions are search gap when reviewing the previous studies. Furthermore,
made in order to maximize the total impact of the budget and the context in which the previous studies have been applied is
in the second stage, maximization of the total score of supported at the company level without regarding national organizations’
R&D projects is considered. goals. This study, however, makes an effort to prioritize tech-
Mokhtarzadeh et al. [17] developed a three-dimensional model nology fields in accordance with the national goals and strategic
for R&D projects portfolio selection which considers three criteria plan when determining the optimum composition of technolog-
including the risk, attractiveness, and competitiveness. Addition- ical projects and the best-suited form of financing technological
ally, their model considers the interdependencies among different projects.
projects in the form of a multi-objective framework with deter- In the presented model, there is an uncertainty on exact values
ministic variables. Alvarado et al. [16] presented a mixed-integer of some parameters (such as maximum allocated capital to the
linear programming (MILP) to optimize the technologies selection TDF) due to the dynamic and imprecise nature of required data.
and apply them in buildings’ energy systems, simultaneously. Therefore, these parameters are assumed to be possibilistic and
This model regards the cost of whole technology’s life cycle, the suitable possibility distribution based upon the opinions of
carbon emissions, real-time energy prices, and demands. decision makers has been estimated for each imprecise parameter
Mashayekhi and Omrani [15] proposed a multi-objective in the form of a trapezoidal fuzzy number. To make the proposed
model for portfolio selection at stock exchange market. Their model more robust against the uncertainty, a fuzzy-robust possi-
model combines DEA with Markowitz mean–variance model by bilistic approach is applied yielding a much more reliable solution
incorporating the risk, return, and efficiency of each portfolio in the face of non-deterministic parameters, which is examined
into the model. Karasakal and Aker [19] proposed two threshold under several computational tests.
M. Shaverdi, S. Yaghoubi and H. Ensafian / Applied Soft Computing Journal 86 (2020) 105892 5
3. Deterministic model Table 2

Pairwise comparison scale (adopted form saaty, 1980).
In this section, the multi-objective proposed model is formu- Scale of pairwise comparison Numeric rating Reciprocal (inverse)
lated to determine the optimal composition of technology fields Equal importance 1 1
and corresponding technological projects which are financed by Moderate importance 3 1
3
TDFs. As previously mentioned, the technological projects could Strong importance 5 1
5
be financed in various ways, such as different types of loan, Very strong importance 7 1
7
joint venture, and direct investment. The decision to determine 1
Extreme importance 9
how the technological projects could be financed and how much 1
9
1 1
Intermediate values 2 4 6
financial resources should be allocated to each project depends 2 4 6
on the return on investment (ROI), risk, and TRL associated with

each technological project. In TDFs, as state-owned institutes, the
decision makers are looking for projects that have met three main 1. Calculate the sum of each column l in the decision matrix:
objectives as far as possible: (1) Maximizing the potential profit;
(2) Minimizing the risk of projects’ failure; (3) Maximizing em- n
ployment. Since these objective functions are in conflict with each ∑
wlm , ∀l.m (2)
other, their integration in the form of a single objective function
l=1
is not possible. Also, the existence of several objective functions
allows managers to give each objective less or more importance 2. Divide each element in the decision matrix column by its
with respect to their desirability. Therefore, we propose a multi- column sum:
objective model for decision-making. The notations of the model wlm
are presented in Appendix A. wlm = ∑n , ∀l.m (3)
l=1 wlm
3.1. Multi-objective linear programming model
3. Calculate the mean of each row:
∑n
In this section, we propose a multi-objective mixed-integer wlm
model for technology portfolio selection. wfinal = m=1
(4)
n
This mean, represents the weight of each risk type
3.1.1. Defining the risk measure
Since our model considers the risk as an important factor Finally, the total risk, composed of market, technology, and finan-
for technology portfolio optimization, at first, a risk measure is cial risks, is calculated by Eq. (5):
defined based on the relation proposed by Mokhtarzadeh et al.
[17]. We define rmj , rt j , rf j as market, technology, and financial Rj = w mj rmj + w tj rtj + w fj rfj (5)
risks with corresponding weight of w mj , w t j , w f j , respectively.
In this way, the risk numbers (rmj , rt j , rf j ), their corresponding
To calculate the risk measure for each technology field, we
use a risk matrix depicted in Fig. 1. The risk matrix has two weights (w mj , w t j , w f j ), and the total risk (Rj ) are in [0, 1] interval.
dimensions: (1) ‘‘Risk probability’’ in the market, technology and
financial risk calculation are about ‘‘how likely is it that a given 3.1.2. Objective functions
project will fail regarding the market, technology and financial as- Our model considers the technology risk and total profit ob-
pects, respectively’’. (2) ‘‘Risk impact’’ is about ‘‘the consequences tained by implementing of each project as well as the social
of the project failure’’. This matrix shows the risk level that is impact of technology development as the concept of job creation.
equal to the ‘‘risk impact’’ multiplied by ‘‘risk probability’’. The We also consider four financing methods of TDFs in our model
high ‘‘risk impact’’ and ‘‘risk probability’’ results in ‘‘high risk’’ in defined as follows:
the matrix (red region), i.e. the upper right-hand of the matrix
shows the high risk (red) and the lower left-hand shows the low 1. Direct investment: in which the TDF invests on the tech-
risk (green). nological projects directly.
One of the best methods for weighting several items is the 2. Joint venture: there is the possibility of partnership with
pairwise comparison matrix proposed by Saaty [31] and this TDFs and other organizations. This partnership brings about
method has been used in the contexts other than AHP and ANP, some advantages, such as financing larger projects with
such as VIKOR [32,33] and PROMETHEE [34]. Therefore, in order greater job creation opportunity, providing the opportu-
to compute the risks’ weights corresponding to each technology nity of supporting more projects by fewer budgets, re-
field for different risk types (market, technology and financial) as ducing administration cost of projects, and finally, reduc-
the weights be accurate and be in the [0,1] interval, we use pair- ing self-financing for TDF, thereby reducing the risk of
wise comparison matrix and the linear normalization method. To investment.
this end, we use Pairwise Comparison Scale shown in Table 2 and
3. Type 1 loan: that is a short-term loan with a smaller
determine the relative importance of each pair in the pairwise
amount of funding and lower profit rate.
comparison matrix. The matrix’ rows and columns show the
4. Type 2 loan: that is a long-term loan with a larger amount
weights of the market, technology, and financial risks, respec-
of funding and a higher profit rate compared to the type 1
tively. For example, if the market risk is much more important
than technology risk, w12 takes 9 and the reciprocal index (w21 ) loan.
would be 19 .
With these in mind, our proposed model is formulated as follows:
The first objective function tries to maximize selecting those
1 w12 w13
[ ]
technological projects with a higher social impact. According to
W = w21 1 w23 (1) the Terrile et al. [21], the reward of technology is associated to
w31 w32 1 the Technology Leverage Factor (TLF). TLF is a measure defined
Once the W matrix is built, we could determine the normal- as the potential leverage that a given technology could have
ized pairwise comparison matrix (Wnorm ) by linear normalization for creating the labor market. Some of the technologies have
method. This normalization method involves three stages: great applications in different fields (such as Nano), so, have
Fig. 1. Risk matrix.
Table 3 Table 4
TRLb
TLF description (Own construction based on [21]). Definition of cjgt (Own construction based on [35]).
TLF jgt level Description TLF jgt coefficient (%) TRLbjgt 1 2 3 4 5 6 7 8 9
Low Component level technology 10 TRLb
cjgt 100% 87% 75% 62% 50% 37% 25% 12% 1%
Medium Subsystem level technology 30
High System level technology 70
Enabling Enabling level technology 100
TRLb
where cjgt is a failure impact coefficient that has a reverse
relationship with the level of the beginning TRL of the tech-
greater TLF and therefore greater job creation potential. Another nological project g in technology field j in the period t and is
important factor which could be considered as a reward for the defined as Table 4. TRLbjgt and TRLejgt are the TRL associated with
first objective function is the job creation by implementing each technological project g in technology field j in the period t at
project. Thus, we define the first objective function as follows: the beginning and at the end of the project (after executing the
G T J project), respectively. ccojgt is ‘‘technology complexity coefficient’’
which takes a value between 0.1 (lowest) to 1 (highest) and is
∑ ∑ ∑
Max Z1 = Emjgt × TLFjgt × θjgt (6)
determined by the technology assessment experts based on the
g =1 t =1 j=1
technical specifications that have been explained in the project
in which Emjgt is the total number of jobs created by the project g BP.
in technology field j in period t. TLFjgt is the ‘‘Technology Leverage Executing the technological projects accompanies by adminis-
Factor’’ coefficient of the technological project g in technology tration costs, and also maintaining the allocated budget is associ-
field j in period t which is determined based on Table 3. θjgt ated with depreciation costs. Additionally, in each period the TDF
is a binary decision variable that takes 1 if the project g in gets the repayments of the loans including the profit of allocated
technology field j starts receiving financial resources in period t loans to the projects. Therefore, the third objective is formulated
and 0 otherwise. to maximize the obtained benefits:
The second objective function minimizes the total risk of tech- J G T
( )
nological projects. We formulate this objective function based on
∑ ∑ ∑ ROIjgt
Max Z3 = αjgt × )−n
the concepts defined by Terrile et al. [21] as well as Mankins [35]
(
j=1 g =1 t =1 1 − 1 + ROIjgt
for technological projects’ risks. Terrile et al. [21] used Technology (( β ′
J jgt ×Prof
G T
) )
Readiness Level (TRL) illustrated in Fig. 2 and the complexity of a ∑ ∑ ∑
k′
+ βjgt ′
given technology field as a proxy for risk. + ×u
k′
They proposed a qualitative model to establish a suitable bal- j=1 g =1 t =1
ance in the technology portfolio. According to the Mankins [35], (( γ ′′
(8)
J jgt ×Prof
G T
) )
the higher is the level at which the beginning TRL of a project
∑ ∑ ∑
k′′
+ γjgt
+ × u′′
would be, the lower is the impact of the project failure. Similarly, k′′
j=1 g =1 t =1
the higher is the complexity of technology, the higher would be
J G T T
the probability of the failure. ∑ ∑ ∑ ∑
The application of TRL in mathematical calculations is seen in − Ajgt × θjgt − h × Bt
different references such as Magnaye et al. [36], Sauser et al. [37] j=1 g =1 t =1 t =1
and Yasseri and Bahai [38]. Therefore, in our model, TRL is used in which αjgt , βjgt , γjgt are the allocated resources for investment,
as a proxy for technological progress of the projects in the Eq. (7) type 1 loan, and type 2 loan affiliated with executing technolog-
and the difference between the TRL in the beginning and end of ical project g in technology field j in the period t, respectively.
the project multiplied by the failure impact and the technology ROIjgt is the return on investment of the project g in technology
complexity in order to minimize the total risk of technological field j at the period t. Prof ′ and Prof ′′ are the interest rate in type 1
projects. Hence, the second objective function with regard to the and type 2 loans, k′ and k′′ are the total number of type 1 and type
TRL and technology complexity is defined as follows: 2 loan installments, and u′ and u′′ are the number of repayments
J G T of type 1 and type 2 loan installments in each period, respectively.
∑ ∑ ∑ TRLbjgt − TRLejgt
Min Z2 = TRLb
cjgt ×( ) × ccojgt × θjgt (7) Ajgt is the administration costs for performing the technological
TRLejgt project g in the technology field j in the period t. h is depreciation
j=1 g =1 t =1
Fig. 2. Technology readiness levels definition [35].
cost of holding the budget, and Bt is the budget available at the considered regarding the interest rates. In other words, this
end of the period t. maximum level of return guarantees that the resources are
This equation is about maximizing the TDF’s benefits. The allocated optimally and the resources with lower interest
investment benefits in each period is determined by the first sec- rates is not allocated to the projects with very high re-
tion of the equation. The second section of the equation is related turn. There is no limitation for maximum level of return
to the repayment of the type 1 loan that is calculated based on the on investment for direct investment or joint venture. This
total number of installments (k′ ) and the number of installments is formulated through the below constraints:
in each period (u′ ). The third section of the equation is related to
ROImin × σjgt ≤ ROIjgt , j = 1, 2, . . . , J , g = 1, 2, . . . , G,
the repayment of the type 2 loan and its calculations is similar to
the type 1 loan. Administration costs and the depreciation costs of t = 1, 2, . . . , T (11)
maintaining the allocated budget are calculated by the fourth and ′
ROImin × λjgt ≤ ROIjgt , j = 1, 2, . . . , J , g = 1, 2, . . . , G,
fifth sections of the equation, respectively. Administration costs
t = 1, 2, . . . , T (12)
are imposed on the TDF when a project is selected and financed,
therefore the cost coefficient is multiplied by the θjgt . ROIjgt × λjgt ≤ ROImax
′
, j = 1, 2, . . . , J , g = 1, 2, . . . , G,
t = 1, 2, . . . , T (13)
3.1.3. Model constraints
Various criteria and constraints as to deciding on appropriate
′′
ROImin × ρjgt ≤ ROIjgt , j = 1, 2, . . . , J , g = 1, 2, . . . , G,
financing method are held by the model: t = 1, 2, . . . , T (14)
• Technology field selection: The technology fields could be ROIjgt × ρjgt ≤ ROImax
′′
, j = 1, 2, . . . , J , g = 1, 2, . . . , G,
put into portfolio composition only if their respective risk is t = 1, 2, . . . , T (15)
lower than the maximum acceptable risk level. The model ′ ′′
presumes that at least one technology field should be se- where ROImin , ROImin , and ROImin are the Minimum accept-
lected in each period. The constraints (9) and (10) guarantee able level of return on investment for direct investment or
′
the aforementioned concept. joint venture, type 1 and type 2 loan, respectively. ROImax
′′
and ROImax are Maximum acceptable level of return on in-
Ejt × Rj ≤ Rmax , j = 1, 2, . . . , J , t = 1, 2, . . . , T (9) vestment for type 1 and type 2 loan, respectively. σjgt is
J a binary decision variable that takes 1 if there is a direct
investment on the project g in technology field j in the
∑
Ejt ≥ 1, t = 1, 2, . . . , T (10)
period t and is 0 otherwise, λjgt is a binary decision variable
j=1
that takes 1 if there is type 1 loan for the technological
in which Ejt is a binary decision variable that takes 1 if the project g in technology field j in the period t and is 0
technology field j is selected in the period t, and 0 otherwise. otherwise, and ρjgt is a binary decision variable that takes
Rj is total risk of the jth technology field and Rmax is the 1 if there is type 2 loan for the technological project g in
maximum acceptable level of risk for technology fields. technology field j in the period t and is 0 otherwise.
• ROI: One of the projects’ approval criteria is their ROI in • TRL constraints for technological projects: Our model con-
the way that there is a minimum and maximum acceptable siders financing only for projects at the technology devel-
level of return on investment for different financing meth- opment phases after the R&D phases (i.e. TRL 3 and above).
ods. These financing methods have different interest rates. Different financing methods have different constraints re-
Therefore, the minimum and maximum levels of return is garding TRL. If the level of beginning TRL of a given project
is between 3 to 5, TDFs can enjoy the direct investment or Bt ≤ Bmax, t = 1, 2, . . . , T (28)

joint venture, and if it is between 3 to 8, type 1 loan could
Bt = Cat , t = 0 (29)
be used, and if it is between 5 to 8, the type 2 loan could be
chosen. Therefore, we add the constraints (16)–(23) to our Xjt ≤ Ejt × M , j = 1, 2, . . . , J , t = 1, 2, . . . , T (30)
model: J
∑
3 × σjgt ≤ TRLbjgt , j = 1, 2, . . . , J , g = 1, 2, . . . , G, Xjt ≤ Bt , t = 1, 2, . . . , T (31)
j=1
t = 1, 2, . . . , T (16) J
TRLbjgt × σjgt ≤ 5, j = 1, 2, . . . , J , g = 1, 2, . . . , G,
∑
Bt = Bt −1 − Xjt + Cat
t = 1, 2, . . . , T (17) j=1
(( β ′
J jgt ×Prof
G
) )
3 × λjgt ≤ TRLbjgt , j = 1, 2, . . . , J , g = 1, 2, . . . , G, ∑ ∑
k′
+ βjgt ′
+ ×u
t = 1, 2, . . . , T (18) k′
j=1 g =1
(( γ ′′
TRLbjgt × λjgt ≤ 8, j = 1, 2, . . . , J , g = 1, 2, . . . , G, J jgt ×Prof (32)
G
) )
∑∑
k′′
+ γjgt ′′
+ ×u
t = 1, 2, . . . , T (19) k′′
j=1 g =1
5 × ρjgt ≤ TRLbjgt , j = 1, 2, . . . , J , g = 1, 2, . . . , G, J G
( )
∑∑ ROIjgt
t = 1, 2, . . . , T (20) + αjgt × ( )−n ,
j=1 g =1 1 − 1 + ROIjgt
TRLbjgt × ρjgt ≤ 8, j = 1, 2, . . . , J , g = 1, 2, . . . , G,
t = 1, . . . , T
t = 1, 2, . . . , T (21) G
∑(
3 × cjgt ≤ TRLbjgt , j = 1, 2, . . . , J , g = 1, 2, . . . , G, αjgt + βjgt + γjgt ≤ Xjt , j = 1, 2, . . . , J ,
)
t = 1, 2, . . . , T (22) g =1
TRLbjgt × cjgt ≤ 5, j = 1, 2, . . . , J , g = 1, 2, . . . , G, t = 1, 2, . . . , T (33)
t = 1, 2, . . . , T (23) in which Camax

˜ t is the maximum level of capital that could
be allocated to the organization at the beginning of the
where cjgt is a binary decision variable that takes 1 if there period t, Cat is the capital that allocated to the organization
is joint venture on the technological project g in technology in the period t, Bmax is the maximum level of the budget
field j in the period t and is 0 otherwise. available at the end of each period, Bt is the budget available
• Technological projects selection and their financing method: in the period t. Xjt is the allocated resources to the jth
Technological projects could be selected only if their re- technology field in the period t.
spective technology field has been selected (constraint 24). • Upper and lower limit for resource allocation: There are
Besides, if a technological field is selected, at least one resource allocation thresholds based on different types of
project in that field should be selected (constraint 25). Each financing (Constraints (34), (40), (42)), and if a project is
project, if selected, could be financed only with one of the selected, all of the required budget to execute it should be
financing methods, which is shown in constraint (constraint financed. (Constraints (35), (36), (41), (43)). Also, a maxi-
26): mum level of partnership is considered for the joint venture
G as to TDF and other partners (constraints (37), (38)). Other
investors consider a minimum project’s success probability
∑
θjgt ≤ Ejt × M , j = 1, 2, . . . , J , t = 1, 2, . . . , T (24)
which should be greater than a specified level (constraint
g =1
(39)).
G
˜min × σjgt ≤ αjgt ≤ Capmax × σjgt , j = 1, 2, . . . , J ,
∑
θjgt ≥ Ejt , j = 1, 2, . . . , J , t = 1, 2, . . . , T (25) Cap
g =1 t = 1, 2, . . . , T , g = 1, 2, . . . , G (34)
σjgt + λjgt + ρjgt = θjgt , j = 1, 2, . . . , J , g = 1, 2, . . . , G, αjgt ≤ L × cjgt × Djgt + 1 − cjgt × Djgt , j = 1, 2, . . . , J ,
( )
t = 1, 2, . . . , T (26) g = 1, 2, . . . , G, t = 1, 2, . . . , T (35)
αjgt ≥ (1 − cjgt ) × Djgt − M × 1 − σjgt , j = 1, 2, . . . , J ,
( )
where M is a sufficiently large number.
• Budget constraints: We assume that in each period a por-
g = 1, 2, . . . , G, t = 1, 2, . . . , T (36)
tion of the maximum state budget is allocated to the TDF
(Eq. (27)), and there is a maximum level of the budget that Ijgt ≤ cjgt × Djgt × (1 − L) , j = 1, 2, . . . , J ,
TDF could hold in hand and transfer to the next period g = 1, 2, . . . , G, t = 1, 2, . . . , T (37)
(Eq. (28)). Additionally, budget allocation to each technology
Ijgt + αjgt ≥ cjgt × Djgt − M × 1 − σjgt , j = 1, 2, . . . , J ,
( )
field is done only when that field is selected (Eq. (30)),
and the total amount allocated to technology fields in each g = 1, 2, . . . , G, t = 1, 2, . . . , T (38)
period cannot be greater than the total budget available in cjgt × q ≤ Pjgt × σjgt , j = 1, 2, . . . , J ,
that period (Eq. (31)). In each period within the planning
horizon, the loans’ profit and the return on investment are g = 1, 2, . . . , G, t = 1, 2, . . . , T (39)
added to the available budget. The budget balance equation Capmin × λjgt ≤ βjgt ≤ Capmax × λjgt , j = 1, 2, . . . , J ,
′ ′
is formulated via the constraint (32). The aggregated amount
t = 1, 2, . . . , T , g = 1, 2, . . . , G (40)
of different types of financing for technological projects in
each field is constrained by the total budget allocated to that Djgt − M × (1 − λjgt ) ≤ βjgt ≤ Djgt × λjgt , j = 1, 2, . . . , J ,
field (Eq. (33)). g = 1, 2, . . . , G, t = 1, 2, . . . , T (41)
Cat ≤ Camaxt , t = 1, 2, . . . , T (27) Cap′′min × ρjgt ≤ γjgt ≤ Cap′′max × ρjgt , j = 1, 2, . . . , J ,
˜
t = 1, 2, . . . , T , g = 1, 2, . . . , G (42) L̃y ≤ x
Djgt − M × 1 − ρjgt ≤ γjgt ≤ Djgt × ρjgt , j = 1, 2, . . . , J ,
( )
x ≤ d̃
g = 1, 2, . . . , G, t = 1, 2, . . . , T (43) Gx ≤ My
Bx = 0
where Cap˜min is the minimum level of funding for direct
investment or joint venture, Capmax is the maximum level Ey ≤ 0
of funding for direct investment or joint venture, Cap′min and Hy = x
Cap′′min are the minimum level of funding for type 1 and type
2 loans, respectively. Cap′max and Cap′′max are the maximum x ≥ 0, y ∈ {0, 1} (45)
level of funding for type 1 and type 2 loans, respectively. where the vectors Q and L and d correspond to the number
Djgt is the total cost that requested for executing the tech- of jobs created, maximum and minimum capacity for financing,
nological project g in technology field j in the period t, cjgt
respectively. The vectors F, W, and A correspond to failure im-
is a binary decision variable that takes 1 if there is joint
pact coefficient, return on investment and administration costs,
venture on the technological project g in technology field
respectively. Additionally, the matrices B, E , G, H and M are the
j in the period t and is 0 otherwise and Ijgt is the allocated
coefficient matrices, and vectors y and x denote the binary and
resources for the joint venture in the technological project
positive variables, respectively. To develop multi-objective robust
g of technology field j in the period t. L is the maximum
possibilistic programming model (MORPP), we consider the con-
shares of the organization in the joint venture (percentage),
Pjgt is the success probability of technological project g in cept proposed by Bairamzadeh et al. [39] which is formulated in
technology field j in the period t and q is the minimum separate objective functions as follows:
success probability of projects for other organization’s joint 3
∑
venture decision. Max obj1 = ωm µm
• Decision variables: The addressed decision variables are bi- m=1
nary or non-negative variables:
Min obj2 = ϕ L(4) − (1 − ∂1 ) L(3) − ∂1 L(4)
[ ]
Xjt , αjgt , βjgt , γjgt , Ijgt , Cat , Bt ≥ 0 and Ejt , θjgt , σjgt , λjgt , ρjgt , + ϑ (1 − ∂2 ) D(2) + ∂2 D(1) − D(1)
[ ]
cjgt ∈ {0, 1} (44) s.t .

(1 − ∂1 )L(3) + ∂1 L(4) y ≤ x
( )
4. MORPP model and solution procedure
x ≤ (1 − ∂2 )D(2) + ∂2 D(1)
Robust Possibilistic Programming (RPP) is an approach to deal Gx ≤ My
with uncertainty when the mathematical model contains fuzzy
Bx = 0
parameters and are not precisely predictable by experts, origi-
nally presented by Pishvaee et al. [27] in the Journal of Fuzzy Sets Ey ≤ 0
and Systems. Hy = x
In operations research, Robust Programming is about the abil-
x ≥ 0, y ∈ {0, 1} , 0.5 < ∂i ≤ 1 (46)
ity of a mathematical model to cope with uncertainties, when
parameters do not follow a specific distribution function or when in which obj1 is a weighted sum aggregation function that inte-
a definite opinion is not available by experts. In this situation, Ro- grates three basic objective functions z1 , z2 , and z3 . ωm (m denotes
bust Programming enables the model to generate a robust solu- the number of objective functions in the basic model) shows the
tion toward even the worst cases. There are two fuzzy parameters weight of each basic objective function based on the DM opinion.
in this study: µm is the satisfaction degree for each zm defined by the linear
(1) The minimum amount of budget assigned for direct in- fuzzy membership functions depicted as follows:
vestment or joint venture, which is highly affected by the 1, if Z1 > Z1PIS
⎧
⎪
country’s budget. Hence, there is no uncertainty in the ⎨ Z − Z NIS
⎪
1
limits of these two types of financing. µ1 (x) = 1
, if Z1NIS ≤ Z1 ≤ Z1PIS
Z1PIS − Z1NIS
(2) The maximum level of the state capital, which is also
⎪
⎪
0, if Z1 < Z1NIS
⎩
dependent on the country’s budget influenced by the fluc-
1, if Z2 < Z2PIS
⎧
tuations of Iran’s oil exports, depicted in Fig. 6. ⎪
⎨ Z NIS − Z
⎪
2
To this end, in study, we applied the RPP which allows the de- µ2 (x) = 2
NIS PIS
Z − Z
cision makers to be benefited from the capabilities of both fuzzy ⎩ 2 2
⎪
⎪
mathematical programming and robust optimization approaches. 0, if Z2 > Z2NIS
, if Z3 > Z3PIS
⎧
⎪ 1
4.1. Multi-objective robust possibilistic programming (MORPP) ⎨ Z − Z NIS
⎪
3
model µ3 (x) = 3
⎪
⎪ Z3PIS − Z3NIS
0, if Z3 < Z3NIS
⎩
In this section we rewrite the base proposed model including
three objective functions, by considering the Pishvaee et al. [27] Afterward, we could define positive ideal solution (PIS) and neg-
RPP model, as follows: ative ideal solution (NIS) for each objective function:
min Zm (x∗m ) , m = 2
{ { }
Max z1 = Qy PIS
Zm =1,2,3 =
max Zm (x∗m ) , m = 1, 3
{ }
Min z2 = Fx
Max z3 = Wx − Ay max Zm (x∗m ) , m = 2
{ { }
NIS
Zm=1,2,3 =
s.t . min Zm (x∗m ) , m = 1, 3
{ }
Eq. (48) minimizes vector implied in (47):

k
∑ wi
F (x) = F1 (x) − β
ri
i=2
subject to:
Fi (x) + wi = εi i = 2, . . . , k
gj (x) ≤ 0 j = 1, 2, . . . , m (48)
In this formulation, wi shows a surplus variable for each of the
epsilon constraints. These surplus variables have added to the
objective function with very low coefficients (β ) so that the objec-
tive improves the function associated to the epsilon constraint(s).
The parameter ri is the range of values for each objective function
Fi (x) , i = 2, . . . , k. The behavior of such coefficients is as a
normalization operator to calculate the sum of the objective
functions.
5. Case study: Technology portfolio selection in IIPF
Our proposed model is developed based on Iran’s TDF: ‘‘Iran

Innovation & Prosperity Fund (IIPF)’’. The whole data used in the
mathematical modeling is taken from IIPF. The primary goal of
establishing IIPF is enhancing advanced technologies level by the
means of supporting commercialization activities, thereby im-
proving economic development level based on national goals. Ac-
cording to the specified enacted strategic national goals,
knowledge-based products and services should be supported as
the substantial technology fields. The eight primary technology
fields to be financially supported are: (1) Biotechnology, (2) Ad-
vanced Materials and Polymer and Chemical Products, (3) Power
and Electronic Hardware, Laser and Photonics (4) ICT & Computer
Fig. 3. Risk components’ weights for 8 technology fields.
Software (5) Advanced Machinery and Equipment (6) Advanced
Drugs (7) Medical Instrument, Appliances, and Equipment (8)
Advanced Products in Other fields.
obj2 controls the ‘‘normalized feasibility robustness’’
( of the solution Thus far, IIPF has provided more than 500 million $ financial
) First and second terms of obj2 , namely (1 − ∂1 )L(3) + ∂1
vector. facilities for about 3000 knowledge-based companies. Since there
L(4) y ≤ x and x ≤ (1 − ∂2 )D(2) + ∂2 D(1) , are related to the is no technology priority in upstream documents for support-
chance constraints, respectively. ϕ and ϑ are penalty units of ing knowledge-based fields, IIPF allocated its financial resources
possible violation on constraints including uncertain parameters. without any specific limit or priority for different technology
The terms (1−∂1 )L(3) +∂1 L(4) and (1 − ∂2 ) D(2) +∂2 D(1) indicate the fields. Furthermore, IIPF has not paid attention to these tech-
difference between the worst-case value of uncertain parameters. nologies’ role and impact in the knowledge-based economy in
allocating the financial resources. Therefore, the IIPF’s financial
4.2. Formulation of the augmented ε -constraint algorithm
resources have not optimally allocated to the knowledge-based
fields, leading to weak performance in terms of selecting the
This algorithm, in order to generate the Pareto front, solves
one single-objective optimization problem for each value of the best-suited resource allocation method. In doing so, our proposed
ε parameter. However, this algorithm usually generates weak so- model is developed to determine an appropriate composition of
lutions because the objective value cannot be improved for some technology portfolio and the optimum resource allocation for
subsequent ε values. To avoid such condition and accelerate the IIPF’s technological projects.
whole process, Mavrotas (2009) proposed an improved version
called the ‘‘augmented ε -constraint algorithm’’. Additional details 5.1. Data
of this algorithm have been addressed by Mavrotas (2009).
A multi-objective optimization problem with k objectives is In the first step, based on our risk calculation method, we ful-
usually expressed as follows: fill the pairwise comparison of technology fields, determine three
risk components (market, technology and financial) weights, and,
risk index of the risk’s components for different technology fields.
Find:
Finally, we determine the risk scores by the relation (4) for each
x = [x1 , x2 , . . . , xn ] technology field. For pairwise comparison and calculating the
that minimizes weights, we used Expert Choice 11 software.
F (x) = [F1 (x) , F2 (x) , . . . , Fk (x)] Since the intended purpose is determining the criteria (i.e. risk
s.t. components) weights rather than comparing technology fields
based on these criteria, the pairwise comparison is carried out
gj (x) ≤ 0 j = 1, 2, ..., m (47) by Export choice software, and in so doing, we ask IPF experts’
To convert the multi-objective model defined in (47) to a single- opinion. Fig. 3 shows the calculated risk weights for 8 technology
objective model by the augmented ε - constraint algorithm, the fields, and Fig. 4 shows the details of the weights for technology
formulation is modified as follows: field 1.
Table 5 order to determine the values of the positive ideal solution (PIS)
Risk calculation for 8 technology fields. and negative ideal solution (NIS) in terms of Z1 , Z2 and, Z3 , we
Technology field wmj rmj wt j rt j wf j rf j Rj solve each model separately and calculate objective function Obj1 .
T1 0.249 0.45 0.594 0.63 0.157 0.72 0.599 Table 7 illustrates the results for positive and negative ideal
T2 0.528 0.49 0.333 0.36 0.140 0.42 0.437 solutions related to each objective. As mentioned in Section 4.1,
T3 0.128 0.45 0.595 0.49 0.276 0.5 0.487
Obj1 is the aggregation of Z1 , Z2 , and Z3 , namely the basic objective
T4 0.259 0.18 0.259 0.32 0.482 0.48 0.360
T5 0.237 0.4 0.449 0.21 0.313 0.56 0.364 functions. Finally, to solve multi-objective problem of maximizing
T6 0.323 0.35 0.446 0.81 0.232 0.72 0.641 Obj1 and minimizing Obj2 we use augmented epsilon-constraint
T7 0.328 0.72 0.414 0.64 0.258 0.49 0.627 method which can be presented as follows:
T8 0.252 0.12 0.422 0.2 0.326 0.2 0.179
3
∑
Max Obj1 = ωm µm
m=1
This software calculates an ‘‘inconsistency index’’ for each s.t .
pairwise comparison matrix in which the inconsistency below the
Obj2 ≤ ε
0.1 means that judgments are rational. The inconsistency index
for all 8 technology fields is put between the interval of [0.004, x∈S (49)
0.05], as shown in the T1 segment of Fig. 4.
where S indicates the feasible region of the problem involving
Hence, the calculated weights are rational and acceptable. the constraints (6) to (44) of the original model in which the
In the next step, we determine risk index by the proposed ma- constraints (27) and (34) (including uncertain parameters) are
trix (Fig. 1), and finally, calculate the total risk of each technology replaced with the corresponding robust counterparts as follows:
field (Table 5).
We consider the planning horizon for 4 periods and 8 tech- Cat ≤ (1 − ∂1 )Camax
˜ t(2) + ∂1 Camaxt(1) , t = 1, 2, . . . , T
˜ (50)
nology fields which the number of the technological projects ((1 − ∂2 )Cap
˜min(3) + ∂2 Cap
˜min(4) ) × σjgt ≤ αjgt j = 1, 2, . . . , J ,
in a given technology field in each period is between [5,13].
As previously explained, four types of financing are considered t = 1, 2, . . . , T , g = 1, 2, . . . , G (51)
which can be classified in 3 groups as follows: As can be observed, the constraint (51) is nonlinear which should
be transformed to linear counterparts. Thus, constraint (51) can
1. Direct Investment and joint venture: for those projects be linearized by introducing the new variable v = σ × ∂ as
with the highest ROI (at least 0.3) and high success proba- follows:
bility (at least 0.7) whose minimum level of financing is
uncertain and is determined by the expert opinion with ((σjgt − vjgt )Cap ˜min(4) ) ≤ αjgt
˜min(3) + vjgt Cap (52)
regarding the investors’ risk attitude. Therefore, we take vjgt ≤ M σjgt (53)
this parameter as a fuzzy parameter.
2. Type 1 loan: for small projects with lower resources re- vjgt ≥ M(σjgt − 1) + ∂2 (54)
quirements in comparison with other projects (between vjgt ≤ ∂2 (55)
0.01 and 5). In fact, the type 1 loan is a short-term loan
The linear version of the Obj2 can be stated as follows:
with smaller installments for the projects with the medium
Obj2 = ϕ1 (1 − ∂1 ) Camaxt (2) + ∂1 Camaxt (1) − Camaxt (1)
[ ]
ROI (0.13–0.2). ˜ ˜ ˜
3. Type 2 loan: for large projects with higher financial re- (56)
+ϕ2 vjgt Cap ˜min(3) − vjgt Cap
˜min(4) − σjgt − vjgt Cap
[ ( ) ]
˜min(4)
quirements (between 2 and 10). This is a long-term loan
with greater installments for the projects with higher risk Values of epsilon ranging from 2.68 to 1315.9 for Obj2 are
and the ROI between 0.16 to 0.35. obtained from the payoff table through ‘‘Augmented epsilon-
constraint’’ method described in Section 4.2. Also, the deter-
Table 6 shows parameter settings of the proposed model. ministic model solved under nominal data in order to be com-
In both loan types, in contrast to the investment, the minimum pared with proposed robust possibilistic model can be shown as
and maximum level of financing is determined with respect to follows:
regulations, and hence, there is no uncertainty in the limits of 3
these two types of financing.
∑
Max Obj1 = ωm µm
The maximum level of state capital that could be allocated
m=1
to the fund in period t (Camax t ) is calculated as a percentage
s.t .
˜
of the government budget which fluctuates with the oil price.
Fig. 5 shows the fluctuations of Iran’s oil export in the recent x ∈ S′ (57)
18 years (2000–2018). With regarding the great variations (rang- ′
where S indicates the feasible region of the original model in-
ing from 20 to 133 B$), the amount of capital allocation cannot
volving the constraints (6)–(44).
be determined exactly. In this regard, we take Camax
˜ t as a trape-
Table 8 shows computational results of solving the multi-
zoidal fuzzy parameter and apply a fuzzy robust possibilistic objective robust model under different epsilon limits and differ-
programming to cope with the uncertainty. ent weights of the objectives. As Table 8 shows, with increasing
epsilon limits, the value of the objective function Obj1 increases,
5.2. Implementation and evaluation which in turn leads to more desirable values for basic objective
functions (Z1 , Z2 , Z3 ).
This section presents the results obtained from presented Fig. 6(a, b, c) shows the pairwise conflicts related to the
model in a real case study. These results are obtained by the basic objective functions. In addition, there is a pairwise conflict
GAMS 24.1.2 using CPLEX solver on a laptop with an Intel core between the objectives of the proposed robust model as well
i5 processor with 1.6 GHz of CPU and 4 GB Ram. (Fig. 6d). As expected, with decreasing the limit held on the Obj2 ,
At first, robust counterpart of the constraints with uncertain as the epsilon-constraint, the value of the Obj1 increase, which
parameters is formulated based on the MORPP model. Then, in clearly shows a sharp conflict between these two objectives.
Fig. 4. Details of risk components’ weights for technology field 1.
Table 6
Parameters.
Parameter Value Unit Parameter Value Unit
(70,100,150,220)
(60,120,180,210)
T 4 Number Camax
˜ t M $
(90,150,200,240)
(50,100,160,200)
J 8 Number Bmax 500 M $
G 10–30 Number ROImin 0.3 %
′
TRLbjgt 1–8 Number ROImin 0.13 %
′′
TRLejgt 3–9 Number ROImin 0.16 %
′
cTRLbjgt 0.12–1 Number ROImax 0.2 %
′′
ccojgt 0.1–1 Number ROImax 0.35 %
TLFjgt 0.1–1 Number Cap
˜min (1, 2, 3, 5) M $
ROIjgt 0.11–0.4 % Capmax 7 M $
′
Ajgt 0.00005–0.11 M $ Capmin 0.01 M $
′
Djgt 0.01–22 M $ Capmax 5 M $
′′
Pjgt 0.003–0.999 % Capmin 2 M $
′′
Emjgt 0.6–5 100 FTE Capmax 10 M $
′
Prof 0.08 % n 3 Number
′
k 6 Number q 0.7 %
′
u 2 Number L 0.5 %
′′
Prof 0.11 % h 0.000365 M $/year
′′
k 15 Number Rj 0.360–0.641 Number
′′
u 3 Number Rmax 0.65 Number
Fig. 5. Iran’s oil export (B$) (2000–2018).
Table 7
Results of positive and negative ideal solutions for basic objective functions.
Objective function Positive ideal solution (PIS) Negative ideal solution (NIS)
310.605 0.375
Z1
{max z1; s.t: x ϵ S} {min z1; s.t: x ϵ S}
−20.779 −0.016
Z2
{min z2; s.t: x ϵ S} {max z2; s.t: x ϵ S}
403.541 −0.103
Z3
{max z3; s.t: x ϵ S} {min z3; s.t: x ϵ S}
Table 8
Objective values and resource allocation.
Objective values and resources allocation for ω = (0.4, 0.3, 0.3)
Epsilon limit Robust objective Objective function values Decision variables
∑ ∑ ∑ ∑ ∑ ∑
obj1 Z1 Z2 Z3 Cat Xjt αjgt βjgt γjgt Ijgt
t j ,t j ,g ,t j,g ,t j,g ,t j ,g ,t
ε ≤ 29.480 0.646 203.285 −13.905 247.463 602.245 685.050 198.000 307.350 179.700 144.000
ε ≤ 243.89 0.692 216.680 −15.097 261.926 640.378 718.531 222.000 304.950 191.400 168.000
ε ≤ 458.29 0.714 223.265 −15.551 272.489 683.258 755.559 222.000 329.000 204.500 168.000
ε ≤ 672.69 0.737 230.770 −15.973 281.240 726.039 790.160 222.000 343.350 224.800 168.000
ε ≤ 887.098 0.758 237.690 −16.302 291.913 768.941 826.935 222.000 368.850 236.000 168.000
ε ≤ 1101.50 0.780 244.525 −16.749 299.856 811.801 860.750 222.000 378.350 260.400 168.000
ε ≤ 1315.90 0.800 251.115 −17.073 310.180 854.650 896.850 222.000 402.150 272.700 168.000
∑ ∑ ∑ ∑ ∑ ∑
Obj1 Z1 Z2 Z3 Cat Xjt αjgt βjgt γjgt Ijgt
ε ≤ 29.480 0.648 201.715 −14.006 247.579 602.249 684.950 198.000 308.350 178.600 144.000
ε ≤ 243.89 0.695 215.095 −15.172 262.830 640.378 719.818 222.000 309.500 188.300 168.000
ε ≤ 458.29 0.718 221.995 −15.664 271.680 683.193 754.500 222.000 324.500 208.000 168.000
ε ≤ 672.69 0.740 229.895 −16.109 279.177 726.131 789.150 222.000 329.650 237.500 168.000
ε ≤ 887.098 0.761 236.600 −16.452 290.521 769.020 825.525 222.000 360.650 242.800 168.000
ε ≤ 1101.50 0.782 243.185 −16.934 297.524 811.843 860.050 222.000 362.250 275.800 168.000
ε ≤ 1315.90 0.802 250.225 −17.316 306.065 854.727 895.061 222.000 374.550 298.500 168.000
∑ ∑ ∑ ∑ ∑ ∑
Obj1 Z1 Z2 Z3 Cat Xjt αjgt βjgt γjgt Ijgt
ε ≤ 29.480 0.643 199.880 −13.742 254.229 602.295 688.100 198.000 352.600 137.500 144.000
ε ≤ 243.89 0.688 213.205 −14.859 270.410 640.305 723.517 222.000 359.850 141.600 168.000
ε ≤ 458.29 0.711 220.755 −15.286 279.549 683.258 758.484 222.000 376.550 159.900 168.000
ε ≤ 672.69 0.733 228.370 −15.831 286.591 726.138 792.556 222.000 379.150 191.400 168.000
ε ≤ 887.098 0.755 235.315 −16.213 296.295 769.020 828.556 222.000 398.550 208.000 168.000
ε ≤ 1101.50 0.776 242.490 −16.634 304.598 811.716 863.554 222.000 409.250 232.200 168.000
ε ≤ 1315.90 0.797 249.925 −17.029 312.373 854.780 897.777 222.000 416.950 258.800 168.000
5.3. Analysis and managerial insight have the largest share of financing. The joint venture enables the
organization to support executing larger projects with greater
Fig. 7 shows the total allocated resources to each technology potential for job creation in widespread fields.
field (a) and the amount of allocated finance to each technology Since the job creation, as the first objective in the basic model,
using different financing methods for the ω = (0.4, 0.3, 0.3) and is of crucial importance for IIPF, we compare technology fields in
ε ≤ 243.89 (the row 2 of Table 8). As seen in Fig. 7a, technology terms of job creation. Fig. 8a shows total job creation by executing
field 4 (ICT & Computer Software) attracts the most financial the projects in each technology field, and Fig. 8b shows the cost
resources, the reason of which goes back to the widespread ap- of job creation in each field; i.e. the lower is the better. As is
plication of ICT. In addition, the ‘‘TLF’’ of this field is greater than seen, technology field 4 has the greatest number of job creation
the other fields, which results in a larger potential of job creation. and attracts the most financial resources for projects executed in
Technology field 3 (Power and Electronic Hardware, Laser and this field. In contrast, technology field 1 (Biotechnology) needs
Photonics) and technology field 8 (Advanced Products in Other the lowest financial support for job creation. In field 1, small
fields) have the second and third ranks in terms of total financing, projects with a high rate of job creation can be implemented
respectively. These fields include a wide range of opportunities in with a relatively small amount of required finance which can be
high-tech industries with a high potential for knowledge-based supported by type 1 loan.
job creation, such as power plants, railroad, construction, cultural The results are consistent with this assumption that small and
industries, etc. medium-sized enterprises (SMEs) are one of the best options to
Fig. 7b shows the percentage of different types of financing be selected in the technology portfolio, especially in developing
for each field. As seen, the share of type 1 loan to finance the countries.
technology fields 1, 2, 5, 6, 7 is greater than the other types IIFP, in addition, thanks to the opportunity created by the joint
of financing since, as mentioned previously, type 1 loan, is a venture, can enjoy smaller self-funding and attracting and lever-
short-term loan and has fewer financing limits. These technol- aging the partners’ resources to reduce the risk of investment.
ogy fields have many small projects (in comparison with other We define two measures in order to compare different financing
projects) that fewer financial restrictive policies are adopted for methods depicted in Fig. 9. Fig. 9a indicates the ratio of job
them. Additionally, they have the advantage of being finished creation to financial resources specified by each financing type.
faster compared to other projects. On the other hand, technology Fig. 9b shows a similar index stating the ratio of total executed
field 3 has much more complex technological projects with great technological projects to financial resources for each financing
capital investment requirements. Therefore, the largest portion type. As is obvious, ‘‘direct investment and joint venture’’ have
of the required financial resources for implementing the projects a better performance as to job creation index. Enjoying joint
categorized in field 3 is satisfied by type 2 loan; as a long-term venture, IIPF can support more projects by less self-funding and
loan with more flexible financing limits. For projects grouped in leverage other partners’ financial resources, giving rise to more
technology fields 4 and 8, direct investment and the joint venture jobs created by this method.
Fig. 6. Pairwise conflicts related to basic objective functions (a, b, c) and robust model objective functions (d).
Type 1 loan sets in the first position in terms of technological (e.g. D̃ = (D(1) , D(2) , D(3) , D(4) ). To generate random values for
projects index and the second position in the job creation index. uncertain parameters under each realization, uniform distribution
This loan corresponds to the smaller projects by which IIPF can is used in the variation interval of Fuzzy parameters, in the
support more projects and also pave the way for job creation way that random values are generated uniformly between two
better than type 2 loan. extreme[ points of ] the related interval (i.e., Lreal ∼ [L(1) , L(4) ] and
Dreal ∼ D(1) , D(4) ). Compact form of the realization model could
5.4. Computational experiments be written as follows:
∗ ∗
To validate the proposed model, we define a deterministic (Qy∗ ) − Z1NIS Z2NIS − (Fx∗ )
Max objReal = µ1 PIS ∗ NIS ∗
+ µ2 PIS ∗ ∗
linear programming model named ‘‘realization model’’ in which Z1 − Z1 Z2 − Z2NIS
the values of uncertain parameters are produced randomly in the (Wx∗ − Ay∗ ) − Z3NIS
∗
corresponding interval of Fuzzy parameters. The model is solved + µ3 PIS ∗ NIS ∗

− ϕ Rrd − ϑ Rrc
under several realizations to simulate the uncertain setting. The Z3 − Z3
comparison of obtained solutions under each realization for both s.t .
deterministic and proposed robust model could properly show Lreal y∗ ≤ x∗ + Rrc
the effectiveness of the robust model.
As stated before, all the uncertain parameters of our model x∗ − Rrd ≤ Dreal
are represented by trapezoidal possibility distribution function Gx∗ ≤ My∗
Fig. 7. Total financing for each technology field (a) and financing of technology fields by financing method (b).
Fig. 8. Total job creation for each technology field (a) and job creation cost in each technology field (b).
Fig. 9. Job creation proportion to financial resources (a) and technological projects proportion to financial resources (b).
Also, we evaluate the robust possibilistic model under ten re-

alizations with different right-hand side (RHS) values of epsilon-
constraint at grid points. Table 10 indicates the comparison of
the deterministic (realization model) and robust performance in
terms of the average and standard deviation under each realiza-
tion while the robust model is solved with various ‘‘epsilons’’.
As can be seen in Table 10, all of the MORPP model pareto
optimal solutions have better performance in terms of average as
well as the standard deviation of the infeasibility penalty values
compared to the deterministic model solutions. Besides, with
increasing the value of RHS in epsilon-constraints, the value of
the average cost and the standard deviation worsen, because the
impact of Obj2 limits is reduced.
Since the optimality robustness objective function (Obj2 ) seeks
to minimize the maximum deviation of the solution vector, re-
ducing the impact of Obj2 leads to the increasing standard devi-
ation.
Fig. 10. Comparing the values of deterministic (realization model) and robust
objectives under various deviation from expected value. 6. Conclusion
So substantial is the role of the ‘‘Technology Development

∗ Funds’’ (TDFs) and private investors to the point where countries’
Bx = 0 infrastructure development highly depend on them. Determining
Ey∗ ≤ 0 optimum technology portfolio is a matter of paramount impor-
Hy∗ = x∗ tance which should be taken into consideration by TDFs in order
to support composition of projects with the lowest risk, highest
x∗ ≥ 0, y∗ ∈ {0, 1} (58) social impact and profits.
One of the challenging issues which TDFs usually confront
in which Rrd and Rrc are the only decision variables that determine
is supporting the innovative firms’ technological activities with
the violation of chance constraints under random realization. ϕ
the effective forms of financing. There are several methods by
and ϑ are the penalty costs of the corresponding constraint viola-
which TDFs can finance technology fields during different stages
tions which are assumed to be equal to the penalty of unsatisfying
of firms’ life cycle, such as different types of loan, joint venture,
the related chance constraints. Table 9 shows the results obtained
direct investment, to name but a few. Due to the importance
under ten realizations in various deviations from the expected of technological innovation and ‘‘knowledge-based economy’’,
value. The objective function of the realization model consists of TDFs explore effective technology financing methods and com-
two terms: average performance and penalty costs. The average mercialization systems continuously. Since different technologies
performance takes a value much lower than the penalty cost in have different impacts on economic development, job creation,
order that of being normalized. Noteworthy, due to the significant and country’s overall position in innovation and competitive-
difference between average performance and penalty costs, the ness rankings, the significance of an optimal resource allocation
objective function of the realization model is appropriate to com- system is getting more attention.
pare the performance of the deterministic model and the robust In this study, a multi-objective mixed-integer model was de-
one. Accordingly, to compare the results of the deterministic and veloped to answer the question of how much financial resource
proposed model under 10 realizations, the values of infeasibility should be allocated to each project and which technology field
penalty (i.e., ϕ Rrd + ϑ Rrc ) is reported in Tables 9 and 10. will give rise to a higher rate of job creation and financial profits.
To evaluate the desirability of the MORPP model, the solutions To this end, the rate of investment, the risk, and TRL associated
offered by the proposed models are executed using several nu- with each project were incorporated into mathematical modeling.
merical tests and the results are reported. Nine test problems To handle the Fuzzy parameters, a multi-objective robust pos-
with various relative deviations from average expected values sibilistic programming (MORPP) was formulated and evaluated
of Fuzzy parameters (% of deviation) within Fuzzy parameters’ under several computational tests. We also showed that the solu-
interval are considered. For each deviation category, ten random tions obtained from the robust possibilistic approach outweighed
realizations are uniformly generated using the random distri- the deterministic model.
bution specified previously. To analyze the performance of the To suggest managerial insights, we defined two measures in
solutions obtained by MORPP, the robust and the deterministic order to compare different financing methods. The computational
model are solved using produced random data corresponding results showed that ‘‘direct investment and joint venture’’ have
a better performance as to job creation index, and type 1 loan
to each deviation category, and their performance is compared
sets in the first position in terms of the number of technolog-
to each other with respect to their average penalty cost and
ical projects index. Future research could focus on developing
standard deviation reported in Table 9 and shown in Fig. 10. As
a framework in which the decisions in terms of financing the
Table 9 shows, the average and standard deviation values of infea-
projects are made in a multi-state dynamic setting. In addition,
sibility penalty in all of the MORPP solutions for various relative
the moratorium period for repayment of loans can be considered
deviations have better performance in comparison with those of when modeling the problem.
the deterministic model. As the relative deviation increase, the
deterministic model and robust model result in a higher expected Declaration of competing interest
cost and standard deviation. Fig. 10 shows that when relative
deviation increase, the cost difference between the models also No author associated with this paper has disclosed any po-
increase. These results suggest that the robust strategy shows tential or pertinent conflicts which may be perceived to have
more efficient solutions for large relative deviations compared to impending conflict with this work. For full disclosure statements
the deterministic model. refer to https://doi.org/10.1016/j.asoc.2019.105892.
Table 9
Performance comparison between the deterministic (realization model) & proposed model under 10 realizations.
RHS of eps-constraint at grid point = 243.89
Realization No. ϕ Rrd + ϑ Rrc
Deviation from expected value of fuzzy parameters
10% deviation 15% deviation 20% deviation
Deterministic model MORPP Deterministic model MORPP Deterministic model MORPP
1 261.426 13.041 381.805 10.698 653.735 7.482
2 153.981 9.081 619.348 13.272 731.773 11.316
3 426.451 10.974 542.770 5.985 350.729 9.504
4 362.950 7.218 355.016 13.044 747.480 5.232
5 446.059 11.739 363.369 12.282 277.235 11.688
6 130.215 6.900 626.945 9.825 1354.337 19.200
7 503.436 4.707 872.175 12.717 890.900 15.456
8 194.591 10.563 474.274 13.635 129.976 15.015
9 147.902 8.049 619.338 10.215 576.232 11.433
10 345.525 13.455 305.028 3.234 921.509 16.923
Average 297.2536 9.5727 516.0068 10.4907 532.0642 12.3249
Standard deviation 137.6592 2.860943 174.7113 3.429122 313.4114 4.338322
Realization No. 25% deviation 30% deviation 35% deviation
1 957.910 4.716 2304.048 53.891 2203.790 4.830
2 930.159 12.804 1051.415 1.938 1677.083 9.333
3 318.657 10.203 826.582 14.220 1863.413 12.051
4 1282.715 12.294 2122.213 19.599 368.107 9.867
5 930.159 12.804 423.500 12.699 2095.996 68.107
6 818.229 9.552 1299.540 10.410 2348.562 9.369
7 507.711 17.778 1763.333 2.028 1479.688 12.222
8 718.211 6.837 1366.654 8.991 1296.539 5.001
9 1009.845 9.045 190.727 16.041 2194.328 14.502
10 1569.985 18.519 1943.156 16.194 904.523 13.428
Average 904.3581 12.4552 1329.117 15.6011 1643.203 15.871
Realization No. 40% deviation 45% deviation 50% deviation
1 1367.526 5.910 936.902 17.289 2267.946 5.871
2 1949.107 6.669 1239.835 26.232 2374.893 28.124
3 1801.523 7.647 3344.638 997.865 3527.383 1180.604
4 1312.860 13.986 2402.516 152.372 2285.428 35.279
5 2434.125 87.350 2479.184 132.410 3172.842 922.700
6 1002.683 12.936 2623.078 276.284 4341.851 2091.698
7 1584.264 13.695 720.810 8.112 2860.751 610.595
8 2814.132 6.156 2809.238 781.351 1799.847 2.676
9 533.099 16.737 2164.490 136.612 1890.893 4.131
10 1650.569 13.587 840.162 14.568 1695.924 5.415
Average 1644.989 18.4673 1956.085 254.3095 2621.776 488.7093
Table 10
The penalty of infeasibility for the deterministic (realization model) and proposed model.
Realization no. ϕ Rrd + ϑ Rrc
Deter. model MORPP model solutions
RHS of eps-constraint at grid points
29.480 243.89 458.29 672.69 887.098 1101.50 1315.908
2360.301 10.143 13.529 164.871 316.212 466.853 618.195 769.542
1 1685.377 13.185 13.185 13.185 13.185 35.251 161.370 287.258
2 2105.160 8.196 77.263 203.381 329.498 455.032 581.151 707.039
3 1858.343 3.621 3.621 3.621 82.703 208.237 334.356 460.244
4 2076.359 6.678 6.678 6.678 129.057 279.698 430.151 582.387
5 1194.858 12.318 12.318 12.318 12.318 12.318 12.318 12.318
6 1853.066 6.483 6.483 6.483 77.405 202.939 329.058 454.946
7 1840.327 6.555 6.555 6.555 6.555 43.685 194.138 346.374
8 2150.001 10.974 122.101 248.219 374.336 499.870 625.989 751.877
9 11.026 11.025 11.025 11.025 11.025 11.025 11.025 11.025
10 2189.509 11.025 11.025 11.025 145.644 296.285 447.627 598.974
Average 1756.757 9.109364 25.79845 62.48736 136.1762 228.2903 340.4889 452.9076
Standard deviation 657.6239 2.984648 38.0151 93.76511 139.9615 188.5688 224.94 268.4306
Appendix A Appendix B. Robust possibilistic programming (RPP) approach
In this section, we briefly explain the robust possibilistic pro-

See Table 11.
Table 11
Definition of notations.
Notations Description
Indices
t Time periods (t = 1, . . . , T)
j Technology field (j = 1, . . . , J)
g Technological projects (g = 1, . . . , G)
Parameters
Rj Total risk of the jth technology field
Camax
˜ t Maximum level of capital that could be allocated to the organization at the beginning of the period t
Bmax Maximum level of the budget available at the end of each period
ROImin Minimum acceptable level of return on investment for direct investment or joint venture
′
ROImin Minimum acceptable level of return on investment for type 1 loan
′′
ROImin Minimum acceptable level of return on investment for type 2 loan
′
ROImax Maximum acceptable level of return on investment for type 1 loan
′′
ROImax Maximum acceptable level of return on investment for type 2 loan
Cap
˜min Minimum level of funding for direct investment or joint venture
Capmax Maximum level of funding for direct investment or joint venture
′
Capmin Minimum level of funding for type 1 loan
′
Capmax Maximum level of funding for type 1 loan
′′
Capmin Minimum level of funding for type 2 loan
′′
Capmax Maximum level of funding for type 2 loan
Rmax Maximum acceptable level of risk for technology fields
n Investment period in venture investment or partnership
q Minimum success probability of projects for other organization’s joint venture decision
h Depreciation cost of maintaining budget
′
Prof Profit percentage in type 1 loan
′
k Total number of type 1 loan installments
′
u The number of installments of type 1 loan in each period
′′
Prof Profit percentage in type 2 loan
′′
k Total number of type 2 loan installments
′′
u The number of installments of type 2 loan in each period
L Maximum shares of the organization in the joint venture (percentage)
M Big M (A sufficiently large number)
ccojgt Technology complexity coefficient of the technological project g in technology field j in the period t
TRLbjgt TRL at the beginning of technological project g in technology field j in the period t
TRLejgt TRL at the end of technological project g in technology field j in the period t
TRLb
cjgt TRL coefficient of the technological project g in technology field j in the period t
TLFjgt Technology Leverage Factor coefficient of the technological project g in technology field j in the period t
ROIjgt ROI of the technological project g in technology field j in the period t
Ajgt Administration cost of the technological project g in technology field j in the period t
Pjgt The success probability of technological project g in technology field j in the period t
Emjgt Expected jobs creation by the technological project g in technology field j in the period t
Djgt Total cost that requested for executing the technological project g in technology field j in the period t
Variables
Bt The budget available in the period t
Cat Capital that allocated to the organization in the period t
Xjt Allocated resources to the jth technology field in the period t
Ejt Binary decision variable that takes 1 if the technology field j is selected in the period t; and 0 otherwise
αjgt Allocated resources for the direct investment in the technological project g of technology field j in the period t
βjgt Allocated resources to type 1 loan for the technological project g in technology field j in the period t
γjgt Allocated resources to type 2 loan for the technological project g in technology field j in the period t
Ijgt Allocated resources for the joint venture in the technological project g of technology field j in the period t
θjgt Binary decision variable that takes 1 if the technological project g in technology field j starts receiving fund in the period t;
and is 0 otherwise
σjgt Binary decision variable that takes 1 if there is a direct investment on the project g in technology field j in the period t and
is 0 otherwise
λjgt Binary decision variable that takes 1 if there is type 1 loan for the technological project g in technology field j in the period
t and is 0 otherwise
ρjgt Binary decision variable that takes 1 if there is type 2 loan for the technological project g in technology field j in the period
cjgt Binary decision variable that takes 1 if there is joint venture on the technological project g in technology field j in the period
gramming model proposed by Pishvaee et al. [27]; then, the basic possibilistic chance constrained programming (BPCCP) is as
Multi-Objective Robust Possibilistic Programming (MORPP) ap- follows:
proach will be elaborated on what follows.
Max z = Qy − ϕ1 L(4) − (1 − ∂1 ) L(3) − ∂1 L(4)
[ ]
Assumed that the compact form of a model could be stated as
− ϕ2 (1 − ∂2 ) D(2) + ∂2 D(1) − D(1)
[ ]
follows:
Max z = Qy s.t .
(1 − ∂1 )L(3) + ∂1 L(4) y ≤ x
( )
s.t .
L̃y ≤ x x ≤ (1 − ∂2 )D(2) + ∂2 D(1)
Gx ≤ My
x ≤ d̃
Bx = 0
Gx ≤ My
Ey ≤ 0
Bx = 0
Hy = x
Ey ≤ 0
Hy = x x ≥ 0, y ∈ {0, 1} , 0.5 < ∂i ≤ 1 (62)
x ≥ 0, y ∈ {0, 1} (59) in which, in contrast to the PCCP model, ∂i is a[ decision vari-

able corresponding to ith chance constraint. ϕ1 L(4) − (1 − ∂1 )
It is assumed that vectors L and d are imprecise parameters.
L(3) − ∂1 L(4) and ϕ2 (1 − ∂2 ) D(2) + ∂2 D(1) − D(1) , the second
] [ ]
According to Pishvaee et al. [27] the basic possibilistic chance
and third terms of the objective function, determine the con-
constrained programming (BPCCP) for the aforementioned model
fidence level of each chance constraint and control the feasi-
could be stated as follows:
bility robustness of the solution vector. In this term, ϕ1 and
Max z = Qy ϕ2 denote the penalty unit of possible violation of the con-
s.t . straints that include uncertain parameter(s). Also, the difference
{ } between the worst case value of uncertain parameters and the
Nec L̃y ≤ x ≥ ∂1 values used in the related chance [ constraints is demonstrated]by
L(4) − (1 − ∂1 ) L(3) − ∂1 L(4) and (1 − ∂2 ) D(2) + ∂2 D(1) − D(1) .
[ ]
{ }
Nec ∂1 x ≤ d̃ ≥ ∂2
References
Gx ≤ My
Bx = 0 [1] J. Wonglimpiyarat, Government programmes in financing innovations:
Comparative innovation system cases of Malaysia and Thailand, Technol.
Ey ≤ 0 Soc. 33 (2011) 156–164.
Hy = x [2] Wonglimpiyarat, Technology Financing and Commercialization; Exploring
the Challenges and how Nations Can Build Innovative Capacity, Palgrave
x ≥ 0, y ∈ {0, 1} (60) Macmillan, UK, 2015.
[3] J. Wonglimpiyarat, Entrepreneurial financing for venture and innovation
where ∂i remarks the minimum confidence level of satisfying the development, Int. J. Foresight Innov. Policy 5 (4) (2009) 234–243, Special
ith chance constraint. According to the Pishvaee et al. [27], Nec Issue.
shows the ‘‘necessity’’ of the related constraints and could be [4] Michael Kahn, Luiz Martins de Melo, Marcelo G. de Matos, Pessoa,
Financing Innovation India, Routledge, 2014.
applied directly to convert the fuzzy chance constraint into its
[5] Michael W. Dickinson, C. Thornton Anna, Stephen Graves, Technology
equivalent crisp constraint. Since using the necessity measure to portfolio management: Optimizing interdependent projects over multiple
satisfy the chance constraint is more meaningful, the necessity time periods, IEEE Trans. Eng. Manage. 48 (4) (2001) 518–527.
measure is applied in order to cope with chance constraint. [6] Oliver S. Yu, Technology Portfolio Planning and Management; Practical
In this study, to model the uncertain parameters, we use Concepts and Tools, Springer Science & Business Media, USA, 2007.
[7] Li Congbo, Liu Fei, Tan Xianchun, Du Yanbin, A methodology for selecting
trapezoidal possibility distribution )which could be defined by its a green technology portfolio based on synergy, Int. J. Prod. Res. 48 (24)
four points: ξ̃ = ξ(1) , ξ(2) , ξ(3) , ξ(4) .
(
(2010) 7289–7302.
The crisp counterpart formulation of the possibilistic chance [8] Farzad Sattari Ardabili, Technology portfolio modeling in hybrid
constrained programming (PCCP) model could be written as fol- environment, Afr. J. Bus. Manage. 5 (17) (2011) 4051–4058.
[9] Hamid Davoudpour, Sara Rezaee, Maryam Ashrafi, Developing a framework
lows:
for renewable technology portfolio selection, Renew. Sustain. Energy Rev.
Max z = Qy 16 (2012) 4291–4297.
[10] Madjid Tavana, Kaveh Khalili-Damghani, Soheil Sadi-Nezhad, A fuzzy group
s.t . data envelopment analysis model for high-technology project selection: A
case study at NASA, Comput. Ind. Eng. 66 (1) (2013) 10–23.
(1 − ∂1 )L(3) + ∂1 L(4) y ≤ x
( )
[11] Farhad Hassanzadeh, Hamid Nemati, Minghe Sun, Robust optimization
x ≤ (1 − ∂2 )D(2) + ∂2 D(1) for interactive multi-objective programming with imprecise information
applied to R & D project portfolio selection, European J. Oper. Res. 238 (1)
Gx ≤ My (2014) 41–53.
Bx = 0 [12] M. Jafarzadeh, H.R. Tareghian, F. Rahbarnia, R. Ghanbari, Optimal selection
of project portfolios using reinvestment strategy within a flexible time
Ey ≤ 0 horizon, European J. Oper. Res. 243 (2) (2015) 658–664.
[13] Madjid Tavana, Mehdi Keramatpour, Francisco J. Santos-Arteaga, Esmail
Hy = x
Ghorbaniane, A fuzzy hybrid project portfolio selection method using data
x ≥ 0, y ∈ {0, 1} (61) envelopment analysis, TOPSIS and integer programming, Expert Syst. Appl.
42 (22) (2015) 8432–8444.
It is worth to mention that the chance constraints in the (PCCP) [14] Ozan Kocadagli, Rıdvan Keskin, A novel portfolio selection model based on
model should be satisfied with a confidence level greater than fuzzy goal programming with different importance and priorities, Expert
Syst. Appl. 42 (20) (2015) 6898–6912.
0.5 (i.e., ∂i > 0.5) which is determined by Decision Maker(s). [15] Zahra Mashayekhi, Hashem Omrani, An integrated multi-objective
Subsequently, the robust possibilistic programming (RPP-I) model Markowitz–DEA cross-efficiency model with fuzzy returns for portfolio
proposed by Pishvaee et al. [27] and developed based on the selection problem, Appl. Soft Comput. 38 (2016) 1–9.
[16] Dagoberto Cedillos Alvarado, Salvador Acha, Nilay Shah, Christos N. [28] B. Zahiri, R. Tavakkoli-Moghaddam, M.S. Pishvaee, A robust possibilis-
Markides, A technology selection and operation (TSO) optimisation model tic programming approach to multi-period location–allocation of organ
for distributed energy systems: Mathematical formulation and case study, transplant centers under uncertainty, Comput. Ind. Eng. 74 (2014)
Appl. Energy 180 (2016) 491–503. 139–148.
[17] Nima Mokhtarzadeh, Saiedeh Sadat Ahangari, Maryam Faghei, Proposing [29] B. Zahiri, R. Tavakkoli-Moghaddam, M. Mohammadi, P. Jula, Multi-objective
a three dimensional model for selecting a portfolio of R & D projects, design of an organ transplant network under uncertainty, Transp. Res. E
in: IAMOT 2016 Conference Proceedings, International Association for 72 (2014) 101–124.
Management of Technology, ORLANDO, FLORIDA, USA, 2016. [30] N. Salehi Sadghiani, S.A. Torabi, N. Sahebjamnia, Retail supply chain
[18] S.E. Schaeffer, L. Cruz-Reyes, Static R & D project portfolio selection in network design under operational and disruption risks, Transp. Res. E 75
public organizations, Decis. Support Syst. 84 (2016) 53–63. (2015) 95–114.
[19] E. Karasakal, P. Aker, A multicriteria sorting approach based on data [31] Thomas L. Saaty, The Analytic Hierarchy Process: Planning Setting
envelopment analysis for R & D project selection problem, Omega 73 Priorities, Resource Allocation, McGraw-Hill International, New York, 1980.
(2017) 79–92. [32] F.K. Wang, C.H. Hsu, G.H. Tzeng, Applying a hybrid MCDM model FOR six
[20] Fusun Kucukbay, Ceyhun Araz, Portfolio selection problem: a comparison sigma project selection, Math. Probl. Eng. 2014 (2014).
of fuzzy goal programming and linear physical programming, Int. J. Optim. [33] T. Thipparat, T. Thaseepetch, An integrated VIKOR and fuzzy AHP method
Control: Theor. Appl. 6 (2) (2016) 121–128. for assessing a sustainable research project, World Appl. Sci. J. 22 (12)
[21] R.J. Terrile, B.L. Jackson, A.P. Belz, Consideration of Risk and Reward in (2013) 1729–1738.
Balancing Technology Portfolios, in: 2014 IEEE Aerospace Conference, Big [34] V. Balali, B. Zahraie, A. Hosseini, A. Roozbahani, Selecting appropriate
Sky, MT, USA, 2014. structural system: Application of PROMETHEE decision making method, in:
[22] Baris Canbaz, Franck Marle, Construction of project portfolio considering 2010 Second International Conference on Engineering System Management
efficiency, strategic effectiveness, balance and project interdependencies, and Applications, 2010, pp. 1–6.
Int. J. Proj. Organ. Manage. 8 (2) (2016) 103–126. [35] John C. Mankins, Technology readiness and risk assessments: A new
[23] Musa Caglar, Sinan Gurel, Impact assessment based sectoral balancing in approach, Acta Astronaut. 65 (9–10) (2009) 1208–1215.
public R & D project portfolio selection, Soc.-Econ. Plan. Sci. 66 (2019) [36] R.B. Magnaye, B.J. Sauser, J.E. Ramirez-Marquez, System development
68–81. planning using readiness levels in a cost of development minimization
[24] M.S. Pishvaee, M. Rabbani, S.A. Torabi, A robust optimization approach to model, Syst. Eng. 13 (4) (2010) 311–323.
closed-loop supply chain network design under uncertainty, Appl. Math. [37] B.J. Sauser, J.E. Ramirez-Marquez, R.B. Magnaye, W. Tan, System maturity
Model. 35 (2011) 637–649. indices for decision support in the defense acquisition process, in: 5th
[25] S.M.J. Mirzapour Al-e-hashem, H. Malekly, M.B. Aryanezhad, A multi- Annual Acquisition Research Symposium of the Naval Postgraduate School:
objective robust optimization model for multi-product multi-site aggregate Acquisition Research: Creating Synergy for Informed Change, 2008.
production planning in a supply chain under uncertainty, Int. J. Prod. Econ. [38] S.F. Yasseri, H. Bahai, System readiness level estimation of oil and gas
134 (2011) 28–42. production systems, Int. J. Coast. Offshore Eng. 2 (2) (2018) 31–44.
[26] M.S. Pishvaee, S.A. Torabi, J. Razmi, Credibility-based fuzzy mathematical [39] Samira Bairamzadeh, Mir Saman Pishvaee, Mohammad Saidi-Mehrabad, A
programming model for green logistics design under uncertainty, Comput. multi-objective robust possibilistic programming approach to sustainable
Ind. Eng. 62 (2012) 624–632. bioethanol supply chain design under multiple uncertainties, Ind. Eng.
[27] M.S. Pishvaee, J. Razmi, S.A. Torabi, Robust possibilistic programming for Chem. Res. 55 (1) (2016) 237–256.
socially responsible supply chain network design: A new approach, Fuzzy
Sets and Systems 206 (2012) 1–20.

1 s2.0 S1568494619306738 Main

Uploaded by

Copyright:

Available Formats

1 s2.0 S1568494619306738 Main

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

1 s2.0 S1568494619306738 Main

Uploaded by

Copyright:

Available Formats

Applied Soft Computing Journal 86 (2020) 105892

Contents lists available at ScienceDirect

Applied Soft Computing Journal

A multi-objective robust possibilistic model for technology portfolio

1. Introduction forms of financial support is a challenging issue which govern-

TP/PP at national organization

TP/PPa at Company level

Stock market portfolio

a TP: Technology Portfolio, PP: Project Portfolio.

3. Deterministic model Table 2

on the return on investment (ROI), risk, and TRL associated with

Fig. 1. Risk matrix.

Fig. 2. Technology readiness levels definition [35].

is between 3 to 5, TDFs can enjoy the direct investment or Bt ≤ Bmax, t = 1, 2, . . . , T (28)

TRLbjgt × cjgt ≤ 5, j = 1, 2, . . . , J , g = 1, 2, . . . , G, t = 1, 2, . . . , T (33)

t = 1, 2, . . . , T (23) in which Camax

cjgt ∈ {0, 1} (44) s.t .

Eq. (48) minimizes vector implied in (47):

5. Case study: Technology portfolio selection in IIPF

Our proposed model is developed based on Iran’s TDF: ‘‘Iran

Fig. 4. Details of risk components’ weights for technology field 1.

Fig. 5. Iran’s oil export (B$) (2000–2018).

corresponding interval of Fuzzy parameters. The model is solved + µ3 PIS ∗ NIS ∗

Also, we evaluate the robust possibilistic model under ten re-

So substantial is the role of the ‘‘Technology Development

Appendix A Appendix B. Robust possibilistic programming (RPP) approach

In this section, we briefly explain the robust possibilistic pro-

x ≥ 0, y ∈ {0, 1} (59) in which, in contrast to the PCCP model, ∂i is a[ decision vari-

You might also like